SnowPro Specialty: Gen AI Question 151
Single answerCortex AnalystA retail company is building a natural-language analytics assistant for business users. They want to use Cortex Analyst so users can ask questions like "What were online sales in EMEA last quarter by product category?" The semantic model must minimize incorrect SQL generation, especially because the SALES table contains fields with ambiguous names such as REGION_CD, CHNL, and NET_AMT. Which action will most directly improve the quality and reliability of Cortex Analyst responses in this scenario?
- A
Create a well-defined semantic model with clear business-friendly names, descriptions, metrics, dimensions, relationships, and curated synonyms for ambiguous fields
- B
Increase the size of the virtual warehouse used by the application so Cortex Analyst can generate more accurate SQL from natural-language questions
- C
Replace the semantic model with a prompt template that includes the raw CREATE TABLE statements for all source tables
- D
Store all business metadata in a vector embedding table and let Cortex Analyst infer joins and metric definitions dynamically at runtime
Show answer and explanation
Correct answer: A
Explanation
The best answer is to invest in a high-quality semantic model. For Cortex Analyst, semantic modeling is the key control point for improving text-to-SQL accuracy, governance, and consistency. In practice, teams should define business-friendly metrics and dimensions, document fields clearly, specify relationships, and include synonyms that reflect how users actually ask questions. This is especially important when physical column names are abbreviated or cryptic. Increasing warehouse size helps execution performance, not semantic accuracy. Likewise, raw schema DDL or embedding-based inference does not replace the explicit business context captured in the semantic model. Snowflake guidance for Cortex Analyst emphasizes using a curated semantic model so natural-language questions can be translated into reliable SQL against governed enterprise data.
- A. Correct.
Correct. Cortex Analyst relies on a semantic model to map business language to trusted data structures and generate accurate SQL. In a scenario with ambiguous column names like REGION_CD, CHNL, and NET_AMT, the most effective step is to provide business-friendly field names, descriptions, defined metrics and dimensions, table relationships, and synonyms. This reduces ambiguity and helps Analyst interpret user intent consistently.
- B. Incorrect.
Incorrect. Warehouse size can affect query performance and concurrency after SQL is generated, but it does not directly improve the semantic understanding of user questions or the correctness of SQL generation. This option reflects a common misconception that more compute improves LLM-driven interpretation quality.
- C. Incorrect.
Incorrect. Providing raw DDL alone is much less effective than a curated semantic model for business analytics use cases. CREATE TABLE statements expose technical schema details but do not define business metrics, friendly names, join intent, or synonyms in a way that Cortex Analyst is designed to use for reliable analytics question answering.
- D. Incorrect.
Incorrect. Cortex Analyst is designed to work from an explicit semantic model, not by inferring the entire analytics layer dynamically from embeddings at runtime. Embeddings can support retrieval scenarios in other GenAI patterns, but they are not a substitute for governed metric and relationship definitions needed for accurate text-to-SQL analytics.